Due to the variable shapes and indeterminate sizes of coating sagging defects, as well as the complexity of the coating inspection environment, traditional visual defect detection methods often suffer from omission and misjudgment issues. To address this issue, this research proposes a visual detection method for detecting such defects based on a deep neural network model, utilizing the ResNet50 deep learning model. The experiment acquired a total of 2500 original image samples through spray coating experiments, subsequently employing image enhancement techniques such as exponential transformation and histogram equalization to augment the dataset, resulting in a cumulative total of 4650 image samples. An in-depth analysis of the network′s composition and parameter settings was conducted. In order to comparatively evaluate the performance of the deep learning network, the MEATS-KNN algorithm was designed as a control experiment for machine vision. The results indicate that the deep learning algorithm achieved higher accuracy compared to the machine vision detection method, with the ResNet50 deep network improving recognition accuracy by 20.3 percent.